The goal of this project is to build and demonstrate an anthropomorphic prosthetic arm and hand that is controlled by cortical output. The human arm and hand have approximately 30 degrees- of-freedom (dot- independent joint rotations) and are very complex mechanical structures. Hands are an example of an advanced evolutionary specialization, which along with binocular vision and bipedal locomotion, led to tool use- a major determinant of human brain development and behavior. Yet, little is known about the neural control of the hand during natural behavior. Regarding active prosthetic hands, there has been a paucity of work on robot hand control and only recently has there been an effort to make a truly accurate functioning hand replica. Primate reach-to-grasp behavior is characterized by four components- reach, hand shaping, orientation and the closing of the fingers around the object. Dexterity, characterized by the active generation of force through the fingertips to maintain stable grasp and/or to manipulate an object, can be considered as an additional component of hand behavior. Given our success in developing an anthropomorphic arm prosthesis, we expect to extract the signals necessary to achieve dexterous prosthetic hand control using activity recorded from populations of single neurons. In our present arm-only control scheme, we have successfully extracted the velocity of the arm from the recorded brain activity. To reach our ultimate goal of dexterous control, we will also need to control wrist orientation, hand shape and finger force application. Since each of these control categories is multidimensional, the overall control problem is very difficult. We will use a number of strategies to address this difficult problem. An interdisciplinary team of neurophysiologists, engineers, statisticians, robotocists and psychophysicists with a strong history of collaboration has been assembled to develop the pieces needed for this project. The project will be led by Andrew Schwartz at the University of Pittsburgh where the prosthetic control will take place. Yoky Matsuoka at Carnegie Mellon will build the highly anthropomorphic robots and behavioral manipulanda. Rob Kass, also at Carnegie Mellon, will develop the extraction algorithms relating neural activity to movement. Marco Santello and Stephen Helms-Tillery at Arizona State University will develop the behavioral tasks using a primate model and then record cortical activity as these tasks are performed. Dr. Soechting, at the University of Minnesota, will provide detailed psychophysical data describing the way subjects exert finger forces to manipulate objects. Peter Allen, at Columbia, will develop automated robotic grasp and finger placement algorithms for the brain-controlled prosthetic hand.